Gamma Contrast

Overview

The Gamma Contrast sample demonstrates how to apply per-image gamma correction using CV-CUDA’s GPU-accelerated gamma_contrast operator. Gamma correction maps each normalised pixel value p to p^gamma, which is widely used to match display transfer functions (e.g. the sRGB standard uses gamma ≈ 2.2) or to adjust the perceptual brightness of an image.

Usage

Basic Usage

Apply standard sRGB gamma correction (gamma = 2.2) to an image:

python3 gamma_contrast.py -i input.jpg

Custom Output Path

Specify a custom output file:

python3 gamma_contrast.py -i input.jpg -o my_gamma_output.jpg

Command-Line Arguments

Argument

Short Form

Default

Description

--input

-i

tabby_tiger_cat.jpg

Input image file path

--output

-o

cvcuda/.cache/cat_gamma_contrast.jpg

Output image file path

Implementation

Gamma Contrast Correction

# This sample uses the ImageBatchVarShape overload (gamma_contrast also accepts a
# plain Tensor with a gamma tensor or a host-scalar gamma/gain).
# Wrap the single HWC tensor as a cvcuda.Image and push it into a batch.
img_cvcuda: cvcuda.Image = cvcuda.as_image(input_image.cuda())
batch = cvcuda.ImageBatchVarShape(1)
batch.pushback(img_cvcuda)

# Build a 1-D float32 gamma tensor, one value per image in the batch.
# A gamma of 2.2 matches the standard sRGB display transfer function —
# values < 1 brighten the image, values > 1 darken/increase contrast.
gamma_np = np.array([2.2], dtype=np.float32)
gamma_tensor = cvcuda.Tensor((1,), dtype=np.float32, layout="N")
cuda_memcpy_h2d(gamma_np, gamma_tensor.cuda())

Key points:

  1. ImageBatchVarShape input: This sample demonstrates the var-shape overload, where a single-image tensor is wrapped with cvcuda.as_image and pushed into a cvcuda.ImageBatchVarShape. gamma_contrast also accepts plain cvcuda.Tensor input/output, with either a per-sample gamma tensor or a host-scalar gamma/gain. The host-scalar overload accepts round=cvcuda.Round.NEAREST (the default) or round=cvcuda.Round.TRUNCATE for integer outputs.

  2. Per-image gamma: The gamma argument is a 1-D float32 tensor with one value per image in the batch, enabling different corrections per image in the same call.

  3. Standard gamma 2.2: A value of 2.2 matches the sRGB display transfer function, darkening mid-tones to compensate for how monitors render brightness non-linearly.

  4. In-place output extraction: The output ImageBatchVarShape contains cvcuda.Image objects; cvcuda.as_tensor converts the first image back to a writable HWC tensor with no data copy.

  5. uint8 passthrough: Because the input is already uint8 RGB8, the operator preserves that dtype and the result can be written directly with write_image.

Expected Output

The output shows the image with gamma-corrected pixel intensities:

../../_images/tabby_tiger_cat.jpg

Original Input Image
../../_images/cat_gamma_contrast.jpg

Output: Gamma-corrected (gamma = 2.2)

CV-CUDA Operators Used

Operator

Purpose

cvcuda.gamma_contrast()

Apply per-image power-law (gamma) contrast correction to an image batch

Common Utilities Used

  • read_image() - Load image as CV-CUDA tensor

  • write_image() - Save gamma-corrected image

  • cuda_memcpy_h2d - Upload the per-image gamma values to the GPU

See Also